What problem does it solve? Under pressure, developers and coding agents often patch the symptom of a bug instead of its cause, producing fixes that silence an error while introducing new defects or leaving identical bugs at other call sites. This Skill enforces a disciplined symptom-to-root-cause workflow so every fix is grounded in evidence rather than guesswork. ## Core Features & Use Cases - Structured Debugging Loop: Drives a bug through reproduction, a single falsifiable hypothesis, evidence gathering designed to kill the hypothesis, root-cause confirmation, minimal fix, sibling-defect sweep, and proof via re-running the reproduction. - Stop-Triggers and Guardrails: Halts when evidence contradicts the hypothesis, reframes after two failed genuine attempts, and skips the ceremony for trivially obvious one-line fixes. - Prevention Trace: Requires a failing-first regression test and a learning note with a concrete guardrail so real bugs cannot silently return. - Use Case: A CI test fails intermittently. Instead of re-rolling fixes, you state "I believe the cause is X; if so, Y will be true; I will check Y by Z", gather contradicting evidence, trace the mechanism end-to-end, fix the actual cause, and pin it with a new test. ## Quick Start Ask the AI to debug the failing test using the debugging skill, stating one falsifiable hypothesis and proving the root cause before editing any code.
